You are on page 1of 18

Inflation without a beginning: a null boundary proposal

Anthony Aguirre
School of Natural Sciences, Institute for Advanced Study Princeton, New Jersey 08540, USA∗

Steven Gratton
Joseph Henry Laboratories, Princeton University, Princeton, New Jersey 08544, USA†
(Dated: February 7, 2008)
We develop our recent suggestion that inflation may be made past eternal, so that there is no initial
cosmological singularity or “beginning of time”. Inflation with multiple vacua generically approaches
a steady-state statistical distribution of regions at these vacua, and our model follows directly from
making this distribution hold at all times. We find that this corresponds (at the semi-classical level)
to particularly simple cosmological boundary conditions on an infinite null surface near which the
spacetime looks de Sitter. The model admits an interesting arrow of time that is well-defined and
arXiv:gr-qc/0301042v2 28 Mar 2003

consistent for all physical observers that can communicate, even while the statistical description
of the entire universe admits a symmetry that includes time-reversal. Our model suggests, but
does not require, the identification of antipodal points on the manifold. The resulting “elliptic” de
Sitter spacetime has interesting classical and quantum properties. The proposal may be generalized
to other inflationary potentials, or to boundary conditions that give semi-eternal but non-singular

PACS numbers: 98.80.Cq, 04.20.Gz, 04.62.+v

I. INTRODUCTION robust and could not be avoided by relaxing simplify-

ing assumptions such as that of homogeneity. Thus the
idea of a temporally finite universe with a singular initial
The ancient philosophical question of whether the uni-
epoch came to dominate cosmology.1 Attention has since
verse is finite or infinite in time, and whether time “had
focused on how this primordial singularity (where some
a beginning”, entered the domain of scientific study dur-
presently unknown theory of quantum gravity presum-
ing the 20th century with the development of generally
ably applies) could give rise to a classical “initial” state
relativistic cosmologies in which the universe is homoge-
that could evolve into the observed universe.
neous and expanding, as observed on the largest accessi-
ble scales. The required “initial” classical state, however, seemed
Chief among these cosmologies, and representing oppo- rather special: the universe had to have been extremely
site answers to the question of the universe’s beginning, flat, and statistically homogeneous (i.e. obey the CP)
were the classical Big-Bang and Steady-State. As FRW on scales larger than the horizon size. The theory of
models, both are based on some form of the “Cosmolog- inflation was devised and widely accepted as a solution
ical Principle” (CP) that the large-scale statistical prop- to this problem of a special initial state: given inflation,
erties of the universe admit spatial translational and ro- a flat, homogeneous universe (with the necessary Gaus-
tational symmetries. The models differ greatly, however, sian scale-invariant density fluctuations) is an attractor.
in their time evolution. In the Big-Bang, the properties That is, within some inflating region of fixed, finite phys-
of the universe evolve in a finite time from a dense, singu- ical size, the CP holds more and more precisely with
lar initial state. In contrast, the Steady-State universe is time. What is perhaps more surprising and less widely
said to obey the “Perfect Cosmological Principle” (PCP) appreciated, however, is that in generic inflation models
in that it admits, in additional to spatial translational the universe also comes to obey, with ever-greater preci-
and rotational symmetries, a time-translation symmetry. sion, the Perfect Cosmological Principle. This occurs be-
Since all times are equivalent, there can be no “beginning cause inflation is generically “semi-eternal”: rather than
of time”, and the universe is infinite in duration. ending globally at some time, inflation always continues
Unlike their philosophical predecessors, the Big-Bang in some regions, and the universe globally approaches a
and Steady-State model were observationally distinguish- quasi-steady-state distribution of inflating and thermal-
able, and astronomical evidence eventually turned nearly ized regions, the statistical description of which becomes
all cosmologists away from the Steady-State. Moreover,
theorems proven within General Relativity showed that
the classical singularity of the Big-Bang cosmology was 1 There have been a number of proposals for avoiding a beginning
of time, but generally these involve either continuing through
the cosmological singularity by invoking quantum gravitational
effects, or modifying GR at the classical level. Our approach aims
∗ Electronic address: to develop a non-singular cosmology without appeal to either
† Electronic address: possibility.

asymptotically independent of time [1]. approximated by

Since inflation generically approaches a steady-state, it
ds2 = −dt2 + e2Ht dx2 + dy 2 + dz 2 .

seems physically reasonable to ask whether the universe (1)
can simply be in an inflationary steady-state, thus avoid-
In this background, bubbles of true vacuum φt nucleate
ing a cosmological singularity or “beginning of time”. In-
at a fixed rate λ per unit physical 4-volume that depends
deed, the possibility of truly eternal inflation was raised
upon the potential U (φ)[12]. The interior of each bubble
soon after inflation’s invention, but no satisfactory model
looks like an open FRW cosmology to observers inside it.
was immediately devised [2], and in subsequent years
For a suitably designed U (φ) (as in open inflation[7, 8]),
several theorems were formulated proving the geodesic
there can be a slow-roll inflation epoch inside the bub-
incompleteness of models globally satisfying conditions
ble so that the FRW regions are nearly flat and homo-
seeming necessary for eternal inflation [3, 4]. These the-
geneous, and have scale-invariant density perturbations.
orems suggested that inflationary cosmologies necessarily
One such region could therefore in principle represent our
contain singularities, but the exact nature of the implied
observable cosmological surroundings.
singularities was obscure.
At any time t, we can derive the distribution of bub-
In a recent paper, we constructed a counter-example bles and inflating region within our comoving volume,
to these theorems by providing a model for geodesically with the aim of showing that the distribution approaches
complete truly eternal inflation [5]. There, we analyzed a steady-state. To avoid complications resulting from
the classical Steady-State model in detail, then extended bubble collisions and the ambiguities in connecting the
our analysis to inflation. Here, we develop the model time-slicings within and outside bubbles, we concentrate
from a different standpoint, focusing on an eternal infla- on the statistics describing the inflating region outside
tion in a double-well inflaton potential, and on the cor- of the bubbles. This region is necessarily unaffected by
responding cosmological boundary conditions. Section II the bubbles’ presence because they expand at the speed
motivates and develops our model, and describes its gen- of light: both its global and local properties depend only
eral features. Various aspects of the model are developed on its initial state at t0 . But we may describe three effects
in subsequent sections: Section III discusses the arrow of of the bubble encroachment upon it.
time in our model, and elucidates the failure of the sin- First, let us consider the inflating region left at time
gularity theorems to forbid our construction; Sec. IV dis- t by bubbles forming since t0 . This region must consist
cusses the cosmological boundary conditions, which are entirely of points each of which does not have a nucle-
specified on a null surface; Sec. V discusses the relation ation event in its past light-cone (PLC) going back to t0 .
of our model to the “antipodally identified” or “elliptic” Denoting the volume of this PLC by Q(t, t0 ), it can then
interpretation of de Sitter spacetime that it suggests, and be shown that such points comprise a volume fraction
briefly discusses quantum field theory in elliptic de Sit-
ter; Sec. VI discusses generalizations and extensions of
−4πλ(t − t0 )
our model. We summarize and conclude in Sec. VII. finf = exp[−λQ] ≃ exp (2)
3H 3

for (t − t0 ) ≫ H −1 [9, 10]. Although finf → 0 for large

II. THE PROPOSAL t − t0 , the spacetime is said to be eternally inflating be-
cause for small λ the physical inflating volume within our
comoving region nonetheless increases exponentially with
In this section we develop an eternally inflating cos- time:
mology based on a double-well inflaton potential U (φ)
with minima at φt and φf , where U (φf ) > U (φt ) ≥ 0. Vinf ∝ finf exp(3Ht) ∼ exp(DHt) (3)
This sort of potential is posited in “old” inflation [6] or
“open” inflation [7, 8]. We will first review semi-eternal for any fixed t0 , with D ≡ 3 − 4πλ/3H 4 .
double-well inflation, then extend this to eternal infla- Second, one can show [9] that at fixed t the distribution
tion, then analyze and address the geodesic completeness of inflating regions about any inflating point is described
of the model. by a fractal of dimension D (that is, the inflating volume
Vinf ∝ r3 finf (r) ∝ rD ) up to a scale of order rB (t, t0 ),
A. Semi-eternal “double-well” inflation
h i
rB (t, t0 ) = H −1 eH(t−t0 ) − 1 (4)

A semi-eternally inflating cosmology naturally arises is the physical radius at t of a bubble nucleated at t0 .
from a generic double-well potential [9, 10]. Consider Third, we may calculate, for a given point in the in-
some large comoving region in which, at an initial time flating region at time, the number per unit time N (r, t)
t = t0 , the energy density is dominated by the inflaton of incoming bubbles of physical radius r. This is
φ, with φ ≃ φf and ∂µ φ ≃ 0. Assume the spacetime to
locally resemble de Sitter (“dS”) space, with (for conve- 4πλr2
N (r, t) = (5)
nience) nearly-flat spatial sections, i.e. with a metric [11] (1 + Hr)4

J+ B. Eliminating the beginning

i0 i0
xi I The above semi-eternally inflating model can be made
eternal by setting the “state” of the universe to be exactly
that state approached by semi-eternal inflation: because
P the statistical properties depend only upon t − t0 , for
Y Q specified conditions at t0 the state at fixed t with t0 →
−∞ is the same as that for t → ∞ with fixed t0 .
X J- The state so obtained has the four basic character-
istics listed above: The spatial sections are exactly flat
II (outside of the bubbles), the bubble distribution (as char-
acterized by the incoming bubble rate) is homogeneous
and independent of time, as is the distribution of inflat-
ing regions about any inflating region, and the bubbles
FIG. 1: Conformal diagram for de Sitter (dS) spacetime. are all statistically identical. The inflating region is a
Each point represents one point in 1+1 D dS, or half of a
fractal of dimension D < 3 on all scales. This means
2-sphere in 3+1 D dS. The left and right (dotted) edges are
identified. The shaded region (region I) is covered by coor- that, although inflating regions exist, the global inflating
dinates with flat spatial sections (spacelike lines) with space- fraction finf is zero, just as the fraction of 3D Minkowski
like infinity at i0 ; the straight, timelike lines represent co- spacetime filled by an infinite 2-plane of finite thickness—
moving geodesics. The null surface J − represents t → −∞. an object of fractal dimension two—would vanish. (The
True-vacuum “test bubbles” (not disturbing the background zero probability that a randomly chosen point is in an
spacetime) are darkly shaded and open toward future time- inflating region accords with the fact [5] that within the
like infinity J + . Also shown are are the light-cones of points PLC of each point there is an infinite 4-volume in which
P and Q in regions I and II that open toward J − , and null bubbles can nucleate toward that point.) Unlike the re-
(“Y”) and timelike (“X”) geodesic segments with an endpoint gion filled by the plane, however, the inflating region is
at xi and crossing J − . statistically homogeneous and isotropic, in that it ex-
actly satisfies the “Conditional Cosmographic Principle”
of Mandelbrot that the statistical description of the in-
for r < rB (t, t0 ) and zero for r > rB . flating region about any given inflating region is inde-
An observer within a bubble can never leave, but will pendent of the inflating region chosen (see Ref. [13] for
eventually be encountered by an encroaching bubble wall a discussion of this and other aspects of “cosmological”
after a typical time τcoll , where τcoll is related to the r- fractals).
integral of Eq. (5) by some transformation between the This model (essentially derived by Vilenkin [9]) would
bubble observer’s proper time τ and cosmic time t. Since seem to have exactly the properties expected of an eter-
this rate depends on t − t0 , a patient and very sturdy nally inflating spacetime, has been straightforwardly con-
observer could in principle discover the global time at structed using the steady-state generated by a semi-
which it formed by counting the frequency of incoming eternally inflating model, and extends to infinite nega-
bubbles. tive cosmic time. Yet the arguments of [3, 4] (discussed
Now, as t → ∞, four things occur. First, the nearly- in more detail in Sec. III) imply that it should be geodesi-
flat spatial sections approach perfect flatness. Second, cally incomplete. This issue can be addressed with ref-
the incoming bubble rate N (r, t) becomes homogeneous erence to the conformal diagram of the model, shown
and independent of time on arbitrarily large physical in Fig. 1. The background inflating spacetime is rep-
scales. To see this, imagine that the rate is inhomo- resented by the lightly shaded region. Each equal-time
geneous at early times because bubble nucleation starts surface is intersected by an infinite number of bubbles
at different times in different regions. But since the im- (indicated by light-cones opening toward J + ), which are
pact rate depends on the initial time only for bubbles concentrated along J + and near i0 .
of radius greater than rB (t, t0 ) ∼ exp[H(t − t0 )], it is The model thus far constructed (defined in the shaded
then homogeneous and independent of time on arbitrar- region henceforth called “region I”) is geodesically in-
ily large scales as t → ∞. Third, and for essentially the complete: all null geodesics (such as “Y” in Fig. 1), and
same reason, the distribution of inflating region around all timelike geodesics (such as “X”) other than the co-
any given inflating region also becomes homogeneous and moving ones, have only a finite proper time (or affine
independent of time on arbitrarily large scales. Fourth, parameter) between a point in region I and coordinate
observers within bubbles lose the information about the time t → −∞ (see, e.g., [5]). Most geodesics thus “leave
“global time” at which they exist (see Eq. (5)), and all the spacetime” to their past, encountering J − , the limit
bubbles become equivalent. Thus the physical descrip- surface of the flat equal-time surfaces as t → −∞. (On
tion of the universe, relative to any fixed length scale the conformal diagram this surface looks like a null cone
such as H −1 , satisfies the Perfect Cosmological Principle emanating from a point at the bottom edge.)
arbitrarily well as t → ∞. Although the spacetime is geodesically incomplete

there is no curvature blowup or other obvious pathol-

ogy at J − , so the spacetime is extendible rather than
singular. One may take the position that this sort of in-
completeness is allowed, since the edge is outside of the
future of any point in the region, and any given thing in F V+
the spacetime was made at some particular coordinate
time t > −∞2 . From this point of view, there is no clear J- X t=const.
reason to reject the model as defined in region I.
It seems quite reasonable, however, to ask instead how ? P
the manifold could be extended, and what could be in the I
extension. We start by extending the manifold to include
J − , which is the boundary of the open set comprising -P II
region I. We shall see, as follows, that on J − the field J-
must everywhere be in the false vacuum. Define φ(λ) as
the field value at affine parameter λ of a non-comoving P
geodesic starting at some arbitrary point xi in region I, F V-
where λ increases away from J − . We know that for
some affine parameter λJ , the geodesic encounters J − ,
and also that if our point is within a bubble, there is also
a finite value λf > λJ at which the geodesic leaves the
bubble and enters the false vacuum φf . Then for λ < λf ,
FIG. 2: Conformal diagram for eternal double-well inflation.
we have
Bubbles are open FRW regions; equal time slices shown as
curved horizontal lines. For clarity we have not included bub-
lim φ(λ) = φf . (6)
λ→λJ ble intersections. Also shown are past light-cones, cut off at
J − , of both a point P and its antipode −P (note that P
If we then require that the field be continuous along any and −P are also reflected across the suppressed two-spheres
geodesic, we then find that the field must be in the false in the 4D case).
vacuum φf everywhere on the surface J − . That is, at
the semi-classical level of description, J − must be an in-
finite null surface of pure false vacuum, through which problem, but with boundary conditions on a null surface;
no bubbles pass. see Sec. IV and the Appendix for more details). Exactly
Let us now examine the global classical structure of the the same argument can be made, however, for any point
background spacetime by momentarily neglecting semi- in region II. Thus specifying classical fields everywhere
classical processes such as bubble nucleations. Then the on J − determines their values everywhere in dS space-
manifold comprised of region I and J − is locally dS (con- time. This means that the conditions found to obtain
stant Ricci scalar R and vanishing Weyl curvature tensor) on J − (by specifying the state in region I and requiring
everywhere, with φ in the false vacuum (φ = φf ). This fields to be continuous) also determine the state in re-
manifold can still be extended past J − , and the obvious gion II and we can extend our model to region II in an
extension is to complete dS spacetime. This is certainly essentially unique way.
a solution compatible with our state at J − , and (as we
will argue in Sec. IV and the Appendix) it seems likely to We may now examine the extension of the model to
be unique. Thus we will take the maximal extension of region II at the semi-classical level by including the bub-
the background spacetime to be full dS spacetime. That ble nucleations. The form of U (φ) indicates that bubbles
is, the non-shaded region of Fig. 1, henceforth called “re- must nucleate3 at a fixed rate per unit physical 4-volume.
gion II” must simply be the rest of dS spacetime. Con- In region I, this led to an asymptotically steady-state
sider now a classical field in the background spacetime bubble distribution which, when made exact, implied
obeying a homogeneous hyperbolic equation. Given any that there are no bubbles passing through J − . Thus in
point P in region I, almost all inextendible non-spacelike region II, though bubbles must nucleate at the required
curves through P intersect J − . Therefore specifying the rate, none must pass through J − .
field values on J − effectively poses a “characteristic ini- The only way this may occur is, in exact symmetry
tial value problem” [15, 16] with a unique solution ev- with region I, to have a steady-state bubble distribution
erywhere in region I (this is the analog of the Cauchy on the flat slices of region II, with the bubbles opening
away from J − . This is illustrated in Fig. 2. Region II is,

2 This was the view taken by investigators of the “cyclic

model” [14] which is geodesically incomplete in a very similar 3 Bubble nucleation has been perhaps most rigorously analyzed in
way, and may be the view taken by the adherents of the classic Ref. [17], and the boundary conditions for bubble nucleation in
steady-state model. our model correspond exactly to those analyzed in their study.

then, a sort of mirror reflection of region I through J − .

Thus the answer to the question of how the model can be
extended, and what lies beyond J − , is that an essentially
identical copy of region I lies in region II, connected by
the infinite null surface J − .
φ=const. }
This completes the basic specification of our model.
The cosmology obeys the CP and PCP in that it ad- O
mits a coordinatization such that all spatial slices are ∆=const.
statistically homogeneous, and statistically the same as
all others. There is no preferred time in this slicing, nor
is there a cosmological singularity: the model is geodesi- V(x)
cally complete. Given the appropriate inflaton potential,
any one of the bubbles could describe our observable sur-
FIG. 3: The decay of: (left) an unstable particle, (right) an
unstable vacuum state. In the particle case, the trajectory
III. THE ARROW OF TIME is classically describable to good precision at early and late
times, but not near the decay (shaded region). Likewise, the
inflaton φ is classically describable at large invariant distance
If we consider all bubbles to expand with time, then ∆ from the nucleation event at O, but not near it (shaded re-
Fig. 2 suggests that while in region I the future is to- gion). This quantum region connects the classically describ-
ward the top of the diagram (“up”), in region II future able field configuration of the bubble interior to that of locally
lies toward the bottom (“down”). This leads us to the dS spacetime.
issue of the cosmological arrow of time (AOT): why does
the time-asymmetric 2nd law of thermodynamics hold
universally, given that fundamental physics is thought to has been often argued, particularly by Penrose [19], that
admit a symmetry (CPT) that includes time-reversal? this connection arises because when gravity is included
There is some consensus that if this question has an an- homogeneity corresponds to an extremely low-entropy
swer, it must ultimately be cosmological, with the time state. We shall assume this correspondence here (and
asymmetry resulting from some qualitative difference be- that the bubble does not begin in some very special state
tween cosmological “initial” and “final” conditions [18] for which the density fluctuations decrease). Under this
that precludes a time-reversal (T-) symmetry of the phys- assumption the physical AOT within any bubble must
ical state in any sub-region of the universe, and hence point away from the bubble walls; globally this means
induces an AOT. that the AOT (where defined) points away from J − .
Although our our model has no “initial” conditions, it As illustrated in Fig. 2, one can therefore indeed draw
does have boundary conditions on J − (discussed in de- timelike geodesics (such as “X”) along which the physical
tail in Sec IV below) and we can discuss the AOT in light AOT reverses, but the reversal always occurs to the past
of them. To do so we must divide the universe into two of any physical observer (all of which are within bubbles),
types of sub-regions: those entirely outside of bubbles, and within a region (the locally dS spacetime) in which
and those partially or wholely within them. Outside of there is no well-defined physical AOT.
the bubbles (or alternatively near enough to J − ) there We have argued that within bubbles the physical laws,
is no local AOT: the description of such a region admits a but not the physical state, admit a symmetry (CPT) in-
time-reversal symmetry. Were we to hypothesize, for ex- cluding time-reversal, while outside of bubbles the laws
ample, an imaginary observer outside of a bubble with its and the state admit time symmetry; but what about the
own AOT (pointing away from the time of its creation), bubble nucleations themselves? Is there not some AOT
that observer would see only T-invariant dS spacetime. telling them “which way” to nucleate, depending upon
The observer could not know whether it moved “up” or which side of J − they are on?
“down” on the conformal diagram, nor if it was in region To clarify this point, consider the process of semi-
I or II, nor if it crossed J − . What the observer is guar- classical bubble nucleation by analogy to the decay of
anteed, however, is that it will eventually be encountered an unstable particle, described by a potential V (x) as in
by, and find itself within, a bubble. Fig. 3. Classically, one allowed trajectory is for the parti-
The bubble interiors are not time-symmetric: within cle to sit at constant x in the false minimum xf (vertical
a bubble, there is a unique time direction in which the line on diagram); a second, equal-energy trajectory is
mean energy density decreases. This direction is away for the particle to “bounce” off of the potential (curved
from the bubble wall/slow-roll inflation epoch, at which line in diagram). A purely quantum description of the
the FRW-region is known to be nearly homogeneous. If system (in the Schrödinger representation) would start,
one bubble is to represent our observable surroundings, for example, with a Gaussian wave function centered at
this direction must correspond to the time direction in xf at some time t = 0, and hence with the position ex-
which the entropy of an isolated system increases. It pectation value hx̂i ≃ xf . With time, the wave function

spreads, and hx̂i increases, eventually approaching the A. The singularity theorems
classical trajectory for hx̂i ≫ xf (we could say roughly
that the particle has “decayed” when hx̂i changes sig- Having examined the AOT, we may turn to the sin-
nificantly from xf .) Note also that the same spreading gularity theorems, for the AOT proves crucial in their
would occur at t < 0 so that this purely quantum de- analysis. Both the older theorems [3] and the newer theo-
scription would be T-symmetric. Now, a semi-classical rem [4] assert that if certain conditions are satisfied every-
description of the system would describe the system clas- where within a spacetime, then not all past non-spacelike
sically at both early and late times, but with a quantum- geodesics have infinite proper time or affine parameter.
mechanical transition connecting the classical trajecto- This indicates that the spacetime is either extendible or
ries at some given time. This transition time is random, contains singularities.
and in an ensemble of such systems (as is required for The older theorem poses four such conditions, of which
a correct probabilistic description) would follow a prob- our model satisfies three, as does region I or II by itself.
ability distribution given by a WKB-type calculation of The fourth condition is that there exists a point P such
the decay rate. Near the transition time the system can- that there is a finite difference in volume between the
not be described classically; we must “shade out” the interior of the PLC of P and that of any point P ′ in
region where only a quantum description is accurate, as the past of P . This is motivated by double-well infla-
in Fig. 3. tion, and claimed as necessary for semi-eternal inflation
(the argument being that if this condition does not hold,
Bubble nucleation can be described semi-classically in then any inflating point will find itself in a bubble at
a similar way. Here, we must attach allowed classical so- the next instant, with probability one). We argue that
lutions of the field equations along some boundary that this condition is not quite necessary: what is required is
represents the nucleation event. To do this in a covari- that there be a finite volume in the region between the
ant way, this boundary must be a surface of zero proper PLCs of P and P ′ in which bubbles can nucleate toward
distance, i.e. a null cone, as shown in Fig 3. A bubble nu- P . Thus if, as in the proof of the theorem, the“past of P ”
cleation “event” is thus comprised of a region (shown as is taken to be the full volume interior to the light-cone
the shaded upper quadrant) where a classical bubble in- pointing in the time direction away from which a bubble
terior solution applies, attached to locally dS regions by nucleated at P would expand, then the fourth condition
a “shaded out” region (within some proper separation applies to region I alone (and correctly implies that it is
squared of the nucleation point) where only a quantum extendible). But it would not apply to the full spacetime
description is valid. To produce a semi-classical descrip- (which is neither extendible nor singular), because the
tion of a spacetime in which nucleation events occur, one relevant part of the light-cone extends only to J − .
must then populate it with these configurations in such a The argument of the newer singularity theorem [4] con-
way that the nucleation sites are randomly situated and sists of the definition of a local “Hubble parameter” H
occur at the correct rate per unit four-volume, while the meant to represent the rate of divergence of neighbor-
classical description, which applies far from the nucle- ing comoving test particles, along with an argument that
ation site, is in accord with the (classical) boundary con- any region in which a suitable average Hav of H is greater
ditions. When the classical boundary conditions are ours, than zero along all geodesics must be geodesically incom-
given on J − , this yields the bubble distribution indi- plete.
cated in Fig. 2. The (semi-classical) boundary conditions We understand Borde at al.’s argument as follows. One
do, then, control the time direction of bubble nucleation, imagines some timelike or null test geodesic with affine
not by introducing some locally-detectable AOT, but by parameter λ in the spacetime in question, then attempts
controlling the allowed global configuration of bubble nu- to construct a timelike vector field uµ (λ) along the test
cleation events. geodesic into its past such that Hav (defined via uµ ) ex-
ceeds zero. It is shown that this can only be achieved
along some finite affine length of the test geodesic, since
There is one final “region” in which we can check the imposed condition rapidly forces uµ towards nullness.
the AOT: the entire universe. Interestingly, we here Borde et al. [4] then take their result to mean that an
find that while each bubble nucleation event is non-time- eternally inflating spacetime is past-geodesically incom-
symmetric, by virtue of the symmetry of the cosmologi- plete.
cal boundary conditions, the statistical description of the We take the hypothetical satisfaction of their averaged
universe does admit a sort of T-symmetry. In Sec. V we Hubble parameter condition for all test geodesics as the
will discuss the possibility of making this symmetry ex- implicit definition of what Borde et al. mean by an eter-
act via an identification on the manifold. This raises the nally inflating spacetime. The logic is that uµ might be
intriguing possibility of having a well-defined (and con- independent of test geodesic and simply be the velocity
sistent among communicating observers) AOT for all ob- field of some set of comoving worldlines in the inflating
servers even while the physical laws and the global physi- spacetime. So what the theorem actually implies is that
cal description of the universe both admit a time-reversal is it impossible to entirely cover a spacetime with such a
symmetry. set of worldlines in a way that allows all test geodesics

cutting these worldlines to obey the Hubble parameter IV. BOUNDARY CONDITIONS AND THE
For illustration, let us consider dS space. We note that
as pure dS is a maximally symmetric spacetime, it does We have seen how extending semi-eternal inflation to
not make sense to regard some parts as inflating and oth- eternal inflation implies particular behavior on the infi-
ers not—one must break the symmetry by adding some nite null surface J − . Here we discuss the converse, de-
other ingredients, and then frame a discussion in terms scribing how the eternal double-well inflation model we
of them. Nevertheless, let us investigate if the geodesics have described can be specified by a particularly simple
defined by having fixed comoving coordinates themselves set of cosmological boundary conditions that are imposed
constitute a uµ field satisfying Hav > 0. It turns out on J − .
that H reduces to the usual ȧ/a in this situation. In the
closed slicing with metric
A. On cosmological “initial” conditions
ds2 = −dt̃2 + H −2 cosh H t̃ dχ2 + sin2 χdΩ2 , 2

The correct specification even of a complete set of
physical laws does not by itself allow prediction of any
the coordinates cover all of dS, so uµ is globally defined. physical system’s behavior; these laws must be sup-
However, H < 0 for t̃ ≤ 0 so Hav goes negative there. plemented by boundary conditions that suffice to fully
Now consider using a single flat or open coordinate patch characterize the system being modeled. The Big-Bang
to define uµ . Hav > 0 here (at least for the appropri- (“BB”) model essentially consists of a set of such bound-
ate choice of time orientation). However, because nei- ary conditions for our observable cosmic surroundings:
ther of these coordinate patches covers the spacetime, at some early time, our region was a hot, dense, nearly
neither does uµ . Furthermore, one may choose an infi- homogeneous and isotropic mixture of particles and fields
nite number of flat coordinatizations of dS, each with a in thermal and chemical equilibrium, in a nearly-flat ex-
different null boundary where the uµ construction fails. panding background geometry with scale-invariant grav-
This makes it clear that the boundary of a given uµ field itational potential perturbations of amplitude 10−5 on
cannot be unambiguously used to define an edge to an the scale of the cosmological horizon.
inflating region. While these initial conditions yield predictions in ex-
We do not believe that the global existence of a suitable cellent accord with astronomical observations, they are
uµ is necessarily the best definition of what is meant by deemed by many to be too special: they seem to com-
an eternally inflating spacetime.4 In particular, equat- prise an extremely small portion of some “ensemble of
ing H with physical expansion entails a tacit assumption all possible initial conditions”, using some (generally un-
that the physical AOT is everywhere in accord with that specified) measure (see e.g. [21]). The theory of inflation
defined by uµ . The model we have proposed could be is widely accepted as a way to broaden the range of al-
covered by a congruence of geodesics (those comoving in low initial conditions by funneling a relatively wide class
two flat coordinate patches covering dS) that would yield of physical conditions into satisfactory BB initial condi-
H < 0 in some regions. However, we have argued that tions [22].
these regions may still be regarded as expanding with re- This is perhaps reasonable as a stop-gap measure as
spect to the physical AOT defined by the cosmological long as the true (pre-inflation) initial conditions are un-
boundary conditions. known. But it does not solve the initial condition prob-
In summary, both singularity theorems postulate con- lem, for not all initial conditions will give rise to infla-
ditions for a region to be “inflating”, and find that such tion [23], nor is it clear that all of those that do give
a region cannot be geodesically complete. However, in- inflation will yield a viable big-bang model [21]. Thus
terpreting these theorems as forbidding eternal inflation one must still assume that suitable conditions emerge
seem to us to require an unwarranted assumption about from the initial singularity; whether this assumption cor-
the global AOT independent of the cosmological bound- responds to a “special” or “generic” condition seems ill-
ary conditions. defined without a description of the singularity, which
would presumably require quantum gravity. But unless
quantum gravity completely alters the way physical the-
ories are applied, it is unlikely to yield a unique initial
4 Indeed the term “eternal inflation” has been used with a vari-
condition; it will still be necessary to supply a quantum
ety of meanings. For example, the recent paper [20] used it to state (or something similar). Even if it is possible to de-
describe models that are eternally inflating to the future, but fine the ensemble of all possible states, the question of
simply geodesically complete to the past (ȧ and/or ä may go how (and why) a state was “chosen” for our universe can
negative there). Such histories, with a globally-defined arrow of only be a metaphysical one, and it seems that, just as in
time, seem physically unrealistic with a typical mechanism for
exiting inflation, since this would render them unstable to the choosing physical laws, we can at best posit a state on
formation of thermalized regions in the putatively eternal early some (possibly symmetrical or aesthetic or philosophical)
phase[2]. grounds, derive its consequences, and compare them to

the observed universe. the fields on J − , create interesting dynamics by forming

The same holds for a non-singular cosmological model bubbles that open everywhere away from J − , and give
such as that proposed in this paper, the basic structure rise to eternal inflation as described in Sec. II and shown
of which is well-described by semi-classical physics, and in Fig. 2.
where there can be no hiding of the cosmological bound-
ary conditions in an initial singularity. We instead spec-
ify a particular set, and analyze them in terms of their C. Quantum mechanical boundary conditions
simplicity, symmetry, and ability to correctly generate
a big-bang like region that can describe our observable Although a fully quantum description of semi-classical
surroundings. phenomena—such as the nucleation of a bubble—is gen-
erally prohibitively complex, we may hope that, be-
cause our set of boundary conditions is so simple at a
B. (Semi-)classical boundary conditions for eternal
inflation semi-classical level, it might be amenable to a simple
quantum formulation, and it would be pleasing to spec-
ify our boundary conditions in an explicitly quantum-
It is conventional to pose boundary conditions for a mechanical way. There are two ways one might think of
set of classical fields by specifying the fields (and gener- doing this.
ally their time derivatives) on a spacelike surface. This
First, one might consider doing usual quantum field
accords with the intuitive idea of specifying the state at
theory on de Sitter space, but with a particular choice
an initial time. An alternative procedure, more relativis-
of vacuum corresponding to our “false vacuum” classical
tic in spirit because it does not assume a particular time
conditions on J − . This might be done by putting the
coordinate, is to specify boundary conditions on a null
field in a Gaussian wavepacket around the false vacuum
surface such as the light-cone of a point. (See the Ap-
at (flat slicing) time t, as in [17], then taking a t → −∞
pendix and Refs. [15, 16, 24] for treatments of the null
limit. One might then assume that this is sufficient to
initial value problem.)
define the quantum state over all of de Sitter space. Al-
Our boundary surface J − is a null surface that, when
ternatively, one could set such an initial state at the t̃ = 0
drawn on the conformal diagram, looks like the light-cone
slice in global (closed) coordinates (7), i.e. on the slice
of a point V− at the bottom (i.e. “at” the conventional
through the “throat” of the dS hyperboloid. This surface
past timelike infinity) and the light-cone of a point V+ at
could then be boosted (infinitely) to become J − . In ei-
the top; see Fig.2). As such, one might expect that one
ther case, the procedure would be analogous to putting
may determine fields (including the spacetime metric)
an unstable particle into a Gaussian in the false vacuum
throughout the space by specifying them on J − . The
at t = 0; away from t = 0 in both time directions, the so-
conformal mapping, however, hides the fact that J − is
lution of the time-dependent Schrödinger equation would
not a fully closed light-cone: there is always a nonzero
evolve away from the unstable “initial” state.
physical volume (of order H −3 or greater) on any space-
With our initial data surface being null, one might at-
like surface bounded by J − . We must therefore ask if
tempt to define the quantum state directly on it. This
specifying fields on J − suffices to fix them everywhere, or
would be conceptually more pleasing, not requiring the
if there is information that may come “though the hole”
limit through a sequence of spacelike surfaces as above.
at V− and V+ . In the Appendix, we argue (using the
Further, a null surface formulation of initial data is rather
Green function for a massive scalar field in dS) that the
elegant [16, 24]. However, unlike for a spacelike initial
fields at V± are irrelevant (as long as they are reasonably
value surface, some points on J − are in causal contact.
well-behaved), as their effect is infinitely diluted.
It is thus less easy to see how to move from a spacelike
We can thus pose boundary conditions for our cosmol-
to a null boundary value surface in QFT, because fields
ogy at the classical level as:
do not commute at lightlike separated points. A sim-
1. There exists an infinite connected null hypersurface ilar effect occurs in so-called “light-cone quantization”
J − of topology R × S 2 , on which the 4-dimensional approaches to QFT in Minkowski space [25], where one
Weyl tensor vanishes and the 4D curvature scalar uses infinite null sheets (or lightfronts) instead of equal
is constant. time slices, and fields again do not in general commute.
The vacuum state for interacting fields is rather easier
2. The inflaton field, with a “double well” potential, to define in the lightfront approach to QFT than in the
is everywhere in the false vacuum on this surface. usual spacelike approach to QFT, and one might expect
3. On this surface, all classical fields are zero or are this to hold in a proper null cone approach to QFT. Un-
in minima of their potentials. This precludes any fortunately, the authors know of no such formulation of
radiation propagating through J − . QFT even in Minkowski space.
Thus while we suspect that our cosmological bound-
This cosmology is classically very dull, as it is just ary conditions may correspond to a the specification
de Sitter space everywhere with no dynamics. However, of a rather simple quantum state, we leave this diffi-
semi-classical bubble nucleations can, without affecting cult problem for future work and here concentrate on

a semi-classical description (though we return to QFT in surfaces. For further details on both the WCH and NBP
Sec. V A). see [28].
The tunneling approach to quantum cosmology [29] ar-
gues that the semi-classical universe emerges via a quan-
D. Discussion of the null boundary proposal tum tunneling event. In the context of models with
open-inflationary potentials the proposal suggests that
The boundary conditions we have proposed are ex- the universe, when first semi-classically describable, is
tremely simple, in line with the view espoused above that most likely to be small and regular, with the field away
one cannot avoid making a specific choice for cosmolog- from the true vacuum; future-eternal inflation can then
ical boundary conditions, and that it is then reasonable ensue. The tunneling is supposed to have occurred out
to make a choice that is as simple and as highly symmet- of a quantum gravitational chaos so severe as to pre-
ric as possible, rather than hope to choose a “generic” clude any space-time description. While our proposal
boundary condition. also leads to inflation from semi-classical boundary con-
ditions, it explicitly avoids any such extreme quantum
gravitational regime.
1. Relation to other proposals Over the years Sakharov has discussed various cosmo-
logical models involving time-reversal (and CPT) invari-
ance [30, 31, 32, 33]. In these models, the universe is
Our proposal is novel in that it requires no treatment
generally symmetric across a singular FRW bounce at
of an initial singularity, and in that boundary conditions
t = 0, at which universe assumes an especially simple
are placed on a null boundary that does not correspond
state; away from this bounce entropy increases in both
to any particular cosmic “time”. It does, however, have
directions of time [31]. In [30] he hypothesizes that phe-
features in common with other proposals for specifying
nomena at t < 0 are the CPT reflections of the phenom-
the state of the universe.
ena at t > 0. In [31, 32] he considers the possibility of
Penrose’s Weyl Curvature Hypothesis (WCH) requires
an infinite chain of further bounces or cycles away from
that initial singularities are constrained to have vanishing
t = 0 in both time directions, and also the possibility
Weyl tensor, corresponding to a low-entropy state. This,
that the minimum-entropy surface might be one of max-
he argues, gives rise to an arrow of time flowing away
imum expansion rather than a singular one [31, 32]. In a
from the initial state. Our proposal is very similar, except
paper primarily about signature change [33] he alludes to
of course that we impose the vanishing of the Weyl tensor
possibility of non-singular time-reversal around a surface
on a lightlike slice running across the universe rather than
of minimal radius in a false vacuum state. Our proposal
at an initial singularity.
clearly has parallels with these ideas. We, however, con-
The Hartle-Hawking No Boundary Proposal (NBP) is
centrate on an infinite non-singular null surface, rather
a prescription for the wavefunction of the universe. It
than a singular spacelike one. Moreover, with the con-
provides Ψ(hij , φ), the amplitude for a given three-metric
cepts of semi-classical bubble nucleation and open infla-
and scalar field configuration on a spatial 3D surface.
tion, we are able to provide a relatively complete physical
The construction is designed to suppress irregular con-
figurations relative to more regular ones, and to thus fa-
vor simple and symmetric states. The notion of temporal
evolution does not appear explicitly. However, one can
2. The horizon problem
associate histories with saddle-point approximations to
the wavefunction, and it turns out that such universes
are often smooth when they are young and small. Thus Inflation was originally conceived as a remedy for
the NBP may be taken to imply that, at the semi-classical troubling issues concerning cosmological initial condi-
level, there should exist a spatial surface in a cosmological tions [34], thus it is useful to compare how such shortcom-
spacetime on which the boundary conditions are partic- ings of the HBB model [22] are dealt with in our model.
ularly simple.5 . Our proposal rather uses a null surface We focus on what is (aside from the singularity problem)
(but see Sec. VI B below). It would be very interesting to perhaps the most vexing of these HBB difficulties: the
develop a quantum prescription in a similar vein to the horizon problem.
NBP which naturally leads to simplicity on certain null The horizon problem is generally framed as fol-
lows: choose two spatially antipodal points on the last-
scattering surface. They are similar in temperature, yet
their PLCs never intersect in a HBB cosmology, so there
5 In the context of models with open-inflationary potentials, the can be no causal connection between them. Inflation is
NBP seems to favor histories in which the scalar field is every- generally thought to solve this problem, because with suf-
where in its true vacuum [26]. The NBP may still be relevant ficient inflation, the PLCs will intersect. But this does
for inflation with the use of an anthropic constraint, or in a
“top-down” approach to calculating quantum probabilities [27]. not suffice for the points to have the same temperature,
Starting off in the true vacuum is however not a problem for because the temperature at each point depends on data
recycling models of inflation, as discussed below. across its full PLC, and there is a portion of each PLC

that does not intersect the other. Taking this into ac- problem (posed in terms of Hubble volumes within the
count, for the two points to have similar properties it boundary condition surface) to approximately the same
must be assumed that there is sufficient inflation, and degree as does inflation beginning in the Planck epoch,
additionally that at inflation’s beginning at time tinf , the and much better than does inflation with an early quasi-
region to the past of the two points is homogeneous on FRW phase.
length scales of order Hinf . But suppose there is a earlier
epoch between some time t and tinf . Then at t the patch
must be homogeneous over a physical length scale of at V. THE ELLIPTIC VIEW
a(t)H(t) The model we have proposed consists of two indistin-
r(t) ∼ H −1 (t), (8) guishable regions, each comprising an eternally inflating
a(tinf )H(tinf ) universe with an AOT (where defined) pointing away
where H(t) ≡ ȧ/a. If a ∝ tα (α = 1/2 for radiation- from an infinite null surface which connects the two re-
dominated expansion) prior to tinf , then r/(H −1 ) ≃ gions. The statistical identity of these universes, along
[a/a(tinf )](α−1) . Thus it is necessary to postulate that with lack of a global physical time orientation, suggests
our region was, at some initial time t0 , homogeneous over some form of an old idea concerning dS, called “the ellip-
[a(tinf )/a(t0 )](3−3α) Hubble volumes. The horizon prob- tic interpretation”8 that would identify the two universes.
lem therefore persists if tinf > t0 and α < 1. There are The idea consists of deeming an event to be rep-
two escapes available within inflation. The first is to set resented not by a single point of a spacetime mani-
tinf = t0 = tpl , so that the expansion is inflationary all fold, but by a pair of antipodal points (defined below).
the way back to the Planck time, before which one can- This corresponds to a topological identification that,
not speak of the expansion at all. The horizon problem applied to our model, identifies regions I and II, and
is then greatly ameliorated, as it must only be assumed maps J − onto itself (the R × S 2 manifold J − becomes
that some regions “emerging” from the Planck epoch are (R × S 2 )/Z2 ). This identification has been subject of
homogeneous over a relatively small (but > 1; see [23]) some previous [36, 37, 38, 39] and recent [40] investiga-
number of Hubble volumes when they are first classically tions.
describable; but whether this constitutes a true solution Pure dS can be represented as a 4D hyperboloid
will be seen only if and when the Planck epoch itself is
−v 2 + w2 + x2 + y 2 + z 2 = H −2 (9)
understood. A second potential escape is to set α ≥ 1, or
(more generally) for ȧ to be non-decreasing, i.e. to have embedded in 4+1 D Minkowski space with metric
past-eternal inflation.
How, then, does the horizon problem look in the con- ds2 = −dv 2 + dw2 + dx2 + dy 2 + dz 2 . (10)
text of our eternal model with no initial time? Let us pose
the problem in a slightly more general manner: when The elliptic interpretation consists of identifying each
specifying the cosmological boundary conditions, must point P with coordinates (v, w, x, y, z) with its antipode
one do so over a region that is very large compared to −P at (−v, −w, −x, −y, −z). In the conformal diagram,
some relevant physical volume such as H −3 ? Posed this this means that points such as P and −P of Fig. 2 are
way, it might seem that because J − is an infinite sur- physically identified; the antipodal map A looks like a
face, a strong horizon problem exists.6 This, however, is vertical reflection and a horizontal shift through one half
not necessarily the case. Imagine J − as the limit of a se- of the diagram’s horizontal extent (Note that although
quence of spacelike slices obtained by boosting the space- this map makes an orbifold of the embedding space, there
like surface given by t̃ = 0 in global coordinates. Because
each such surface has volume 2π 2 H −3 , we might also at-
tribute to J − the same finite invariant volume.7 Thus
our construction would seem to ameliorate the horizon −v1 v2 +w1 w2 +x1 x2 +y1 y2 +z1 z2 [11]. In general −∞ < D < ∞,
whereas for two points on J − , |D| ≤ 1; they may thus be con-
sidered “close”. The relation between D and geodesic distance is,
unfortunately, not always defined. The points are timelike or null
separated, respectively, for D > 1 or D = 1. For |D| < 1 points
6 This is apparently the case, for example, in the cyclic model [14], are connectable by a spacelike geodesic of length cos−1 (D) < π;
where if the cyclic behavior is to continue indefinitely into the but for D ≤ −1 there is no geodesic connecting P1 and P2 , al-
future and past, it seems necessary to place cosmological bound- though one does connect P1 to the antipode P¯2 of P2 because
ary confitions on an infinite spacelike surface. This specification D(P1 , P¯2 ) = −D(P1 , P2 ). Under the identification of antipodal
accords the same properties to points that are arbitrarily distant, points (see below), any P1 and P2 are null-separated or spacelike-
and causally disconnected. separated by a geodesic distance < π/2H.
7 One can also consider other ways of taking a limit to J − , but 8 The etymology of this term is obscure to the authors. The el-
this one seems most in accord with the symmetries of dS. An- liptic view applied to spatial sections of dS was discussed first
other approach to the problem is to rather consider some dis- by de Sitter, who preferred it to the (now) conventional “spher-
tance measure between any two points on J − . A dS-invariant ical” view; and cited a letter from Einstein voicing the same
quantity, in units where H = 1, between points P1 and P2 in preference [35]. Antipodally identifying in space and time was
the embedding space (see Eq.(9) below) is given by D(P1 , P2 ) = discussed in fond detail by Schrödinger [36].

tipodal identification, which holds in pure elliptic de Sit-

ter space (edS), is as follows. An immortal observer in
dS can only eventually “see” (i.e. be connected to via
a non-spacelike geodesic emanating into the past of the
observer) half of the space, the rest being hidden behind
J- -P the observer’s dS event horizon. Under the antipodal
identification, however, this “hidden” space is exactly the
P same as the “visible” space. Likewise, the space behind
the observer’s particle horizon (the space not reachable
J- by non-spacelike geodesics emanating from an observer
toward its future) is the same as the space within it. In
this precise sense, edS has no horizons. The notion that
in edS each observer has “full information” about the
space has been a prime motivation for the study of the
FIG. 4: Conformal diagram of dS including Killing vectors space [36, 40].
(arrows), a point P and its antipode −P , along with their Along with these appealing attributes however, and
light-cones. The causal diamond of an observer following a like our cosmological model, edS has some unconven-
geodesic “X” (comoving in the global coordinatization (7)) is tional temporal features that may or may not be desir-
the interior of the dark diamond, the antipodal copy of which
able. Because A includes time reversal, the spacetime is
is the light diamond.
non-time-orientable: one cannot continuously divide non-
spacelike vectors into two classes which can be labeled
are no fixed points in the dS hyperboloid, leaving the “future” and “past”. Now, one may take the view [36]
identified space a manifold). that since physics is essentially time-reversible, this poses
At the classical level, the identification can be enforced no fundamental problem. Non-time-orientability does,
by demanding that all fields in dS are symmetric or anti- however, have implications for quantum mechanics (see
symmetric under A, and that all sources have an accom- Sec. V A and [38, 39]). In addition, while physics may
panying antipodal copy. Parikh et al. [40] have argued be time symmetric, our physical world manifestly is not,
that charge conjugation should be added to A, and that and this must be confronted in a cosmological model.
the combination represents CPT. Indeed, in classical field The identification of points near “past” infinity with
theories at least, A-symmetry automatically entails op- those near “future” infinity also raises the specter of
posite charges at antipodal points. Consider, for exam- closed timelike curves (CTCs) and their accompanying
ple, a complex scalar field, satisfying φ(−P ) = φ(P ). paradoxes. The identification does not allow any self-
Then the global time derivative of the field at the an- intersecting timelike curves in perfect dS because the full
tipode −P is √ minus that at P , and hence the charge light-cones of P and −P never intersect (the spacetime
density (q/2i) −g(φ∗ ∂ 0 φ − φ∂ 0 φ∗ ) is opposite. How- obeys the strong causality principle of [41]). For the same
ever, in 3+1 D at least, we do not recover the PT part reason, no observer can see both an event and its antipo-
of their argument. A particle at P with 3-momentum dal copy. Note, though, that perturbations of dS that
vector p = (px , py , pz ) has an antipodal copy at −P . By tend to make the conformal diagram “taller” [42] would
an argument like that of Parikh et al., parallel trans- allow timelike curves from a point to traverse the dS hy-
porting the copy’s trajectory back to P takes the mo- perboloid to the point’s antipode.
mentum to (−px , py , pz ). This looks like parity followed These temporal features may not be what we are ac-
by a rotation of π about the x-axis. The same proce- customed to, but perhaps we should not be surprised
dure, however, takes a small displacement x = (x, y, z) that any cosmology based on globally dS spacetime has
to (x, −y, −z). The difference in sign arises because the an “unconventional” AOT. It is well known, for example,
3-momentum suffers an “automatic” time reversal under that while a patch of dS admits a static coordinatization,
A, whereas the 3-displacement does not. Thus the orbital it admits no global timelike Killing vector; as shown in
angular momentum l = x × p transforms from (lx , ly , lz ) Fig. 4, the Killing vector is timelike only in half of the
to (−lx , ly , lz ), just as the 3-momentum does. But under diagram, and moreover points toward the top of the di-
parity P, x → −x and p → −p, while under T, x → x agram only in half of the region in which it is timelike,
and p → −p. Therefore under PT (along with possi- reversing completely between antipodal points. (In fact,
ble rotations), p and l transform oppositely. Only under the Killing vector field maps onto itself under A).
T (with rotations) alone do they transform likewise. So Our model has much the behavior one would want in
unfortunately we cannot concur with the beguiling idea a cosmology based on edS. The physical AOT is defined
that CPT conjugate events occur at antipodally conju- only inside the bubbles, which may intersect (and hence
gate points in an elliptic universe. Rather, we must settle compare time orientations) only within region I or region
for a sort of CT conjugation between processes at antipo- II. The two regions are separated by locally dS space-
dal points. time, where physics is fully time symmetric. Under the
A second interesting (and favorable) feature of the an- antipodal identification, the regions are equated, and the

physical AOT is consistent everywhere that it can be a brief summary.

compared by two physical observers. Only an imaginary Consider a massive free scalar field φ(x) obeying some
observer that could travel “back in time” to leave a bub- wave equation, for which we would like to construct a
ble, pass through J − , and encounter another bubble, QFT in edS by defining a Hilbert space of states (in-
could see that its own AOT agreed only with that of one cluding a vacuum state |0i), and the two-point func-
of the two bubbles. (In fact, it would seem that in a non- tion h0|φ(x)φ(y)|0i. The latter can be decomposed
time-orientable manifold the physical AOT must either into a commutator D(x, y) and an anti-commutator (or
be undefined in some regions, or must suffer a reversal “Hadamard”) function G(1) (x, y). Under the antipodal
along some surface. Thus a construction something like identification, we would expect both D and G(1) to be
ours may well be necessary in any cosmology based on symmetric in some sense under the exchange of x and/or
edS.) y with its antipode. How, then, might we define such a
However, an antipodally identified version of our model QFT? There are a number of ways, none of which seem
does not quite share all of the desirable properties of entirely satisfactory.
edS. The bubbles, with a larger curvature radius than First, one might just pick a particular “antipodally
the embedding space, allow the connection of antipodal symmetric” vacuum state of full dS. Indeed, taking the
(and hence identified) points by a timelike curve. These α → ∞ limit of the “α-vacua” appropriate for dS [43]
self-intersecting timelike curves (SITCs) are not however does yield an A-symmetric G(1) . However, this does not
CTCs of the usual grandfather paradox sort. To follow have the usual short distance behavior of the Minkowski
such a SITC an imaginary observer would, for part of 2-point function. In addition, the commutator D is inde-
its journey, have to travel backward in (bubble) time. pendent of the state chosen, and has no antipodal sym-
Moreover, when the two branches of the SITC meet, they metry. (One might also hope to find an A-symmetric
have opposite time orientation as defined by an affine non-vacuum state with the correct short distance behav-
parameter along the curve. One might avoid these SITCs ior, but this would still have the wrong commutator.)
if the bubbles have a smaller curvature radius than the A second approach would be to try to build an an-
background space (as discussed below in Sec. VI). In tipodally symmetric vacuum in a full dS background, by
this case, however, horizons would return, because there choosing a (global) time coordinate and decomposing the
would be regions outside of an observer’s horizon that fields into global positive-frequency modes that are an-
are not identified with any region within the horizon. tipodally symmetric. The Fock vacuum would then as
While the elliptic interpretation is complicated by the usual be the state destroyed by all annihilation opera-
presence of bubbles, the background space of our model is tors. The problem with this approach is that any an-
pure dS, and does benefit from the elliptic interpretation; tipodally symmetric mode turns out to have vanishing
all points on J − , for example, would be connectable by Klein-Gordon norm when integrated over a Cauchy sur-
geodesics and have a maximal spacelike geodesic separa- face for all of dS [38], and the Fock construction breaks
tion π/2H, and the volume of the boundary condition down.
surface would be halved. A third approach employed in the literature [38, 39,
The elliptic interpretation was suggested by a symme- 40] is to abandon the hope of a Fock vacuum, and to
try in the statistical description of the bubbles, and it is simply enforce the antipodal symmetry at the level of
interesting to ponder the connection between this statis- the Green functions, by writing antipodally symmetrized
tical symmetry and A. Classically, we may have an en- versions of the fields, and computing the resulting two-
semble of systems each of which is not A-symmetric, even point functions in terms of the two-point functions of
while the statistical properties of the ensemble are. With- unidentified dS. One problem with this approach is that
out bubble nucleation, our universe (dS) is A-symmetric. it becomes somewhat unclear why the anticommutator
The boundary conditions which determine the bubble should take this value, as the construction seems to lack
distribution are A-symmetric, therefore it seems neces- an underlying quantum-mechanical motivation. Another
sarily true that the statistics of the bubble distribution problem is that while the anticommutator so obtained
are A-symmetric. Given quantum mechanics, however, is antipodally symmetric, the same procedure yields a
it is not clear how to relate a single member to the sta- commutator that vanishes identically.
tistical properties (given by the wavefunction) without A fourth approach is to define the vacuum in terms
bringing in measurement theory, and we will leave the of a mode decomposition over only part of dS (such as a
question aside for future consideration. “causal diamond”[11]), where the modes can be consis-
tently positive or negative frequency (so that a Fock rep-
resentation exists), then provide a prescription for defin-
A. QFT in edS ing correlators between any two points in the space in
terms of correlators within this region. This approach
The non-time-orientability of edS also makes quantum turns out to be promising; for one or both points in the
field theory on the space rather more subtle than in usual causal diamond or its antipodal copy, and for a particular
dS. We hope to treat this subject in some detail in a choice of state (a mixed thermal one at twice the usual
forthcoming paper (see also [38, 39, 40]). Here we sketch de Sitter Hawking temperature), both D and G(1) can be

made to have the correct symmetry. The problem arises on J − . Since the lowest vacuum state must be posi-
when both points are outside of the causal diamond and tive for this scenario to work, it also has the potential
its copy; in this case the commutator turns out to van- to connect with the presently observed positive vacuum
ish for timelike separated points, and not for spacelike energy.
separated points. It is unclear whether this makes sense. A somewhat similar scenario can be realized using
In short, defining a satisfactory QFT in edS is rather “chaotic inflation” potentials such as V (φ) = λφ4 + V0 .
difficult; the difficulties stem primarily from the commu- Here we set the inflaton to rest at φ = 0 on J − , and
tator function, because it is not symmetric under time- require V0 > 0 so that the space is locally dS near J − .
reflection, while the anti-commutator is. It is then diffi- This is, again, a stable configuration, yet regions of large
cult for both functions to be symmetric under A, which potential may “nucleate” in this dS background as highly
includes time-reflection. It is possible [40] that string improbable quantum fluctuations in the field. If such a
theory in edS will make more sense than in dS. It is also region nucleates at high enough φ, it would provide the
conceivable that the elliptic view could emerge from a potential seed for the sort of eternal inflation envisioned
correct quantum treatment of dS, and that the described by Ref. [46], in which quantum fluctuations in φ domi-
troubles stem from doing QFT on a fixed background not nate over classical rolling so that inflation become eter-
included in the dynamics. This would be an interesting nal. While the structure of the resulting region becomes
issue to pursue in string theory or other theories of quan- extremely complicated, it can again be contained within
tum gravity. For now we must leave it there, and return a light-cone of some point in the original (dS) embedding
to eternal inflation. space, so that the global structure of the universe (filled
with such light-cones) is still understandable, and again
looks like Fig. 2.

We have constructed our eternally inflating universe

using an “open inflation” double-well potential and de-
B. Different boundaries
manding that the inflaton φ rest in the false vacuum
everywhere on the infinite null surface J − . But these
choices are not unique, and the general principles of our We were led to the null surface J − by constructing an
construction can be extended to models employing differ- eternal model with a time-translation symmetry. But the
ent potentials, or different boundary condition surfaces. same sort of boundary conditions we apply on J − could
also be applied to a spacelike initial surface, such as the
t̃ = 0 surface in the global coordinatization (7) of dS.
A. Different inflaton behaviors Such boundary conditions might be closely tied in to the
Hartle-Hawking NBP, as discussed above. The universe
A simple change in our model can be induced by leav- now has a preferred (and initial) time, and constitutes
ing V (φ) fixed, but demanding that φ rest in the true, “semi-eternal” inflation in both the t̃ < 0 and t̃ > 0
rather than false, vacuum on J − . Although the field is regions. The same arguments concerning the AOT apply
now in a (positive) stable vacuum, bubbles of false vac- here: it is undefined near t̃ = 0 (outside all bubbles),
uum may still be able to nucleate [44]. Assuming this and defined in bubbles, pointing away from t̃ = 0. If the
indeed occurs, the effect of each bubble can be enclosed initial surface maps onto itself under the antipodal map,
within a light-cone, and the distribution of these light- we can apply the antipodal identification to the universe.
cones is essentially the same as for bubbles of true vac- This model bears a stronger resemblance to the earlier
uum, so all of the arguments of Sec II go through. Now, ideas of Sakharov [30, 31, 32, 33] than does our proposal
within these bubbles of false vacuum new bubbles of true using the null J − .
vacuum can form, one of which could describe our ob- Spacelike surfaces are easily deformable into other
servable universe. At late (bubble) times each bubble spacelike surfaces, whereas the same is not true for null
interior would approach dS, dominated by the true vac- surfaces (see e.g. [47]). Thus spacelike surfaces are in
uum (presumably of the magnitude we currently appear some sense less constrained than null ones and thus
to observe), in which false vacuum bubbles may nucleate maybe less appropriate in any attempt, such as ours,
(and so on, ad-infinitum). This is a realization of the to specify cosmological boundary conditions in the most
“recycling universe” of [45]. Because bubbles are nested economical way. In addition, the resulting universe would
infinitely deep, the structure of this universe appears ex- not obey the Perfect Cosmological Principle, having a
ceedingly complicated. But the global structure at the preferred time at t = 0. As discussed above, we also
outermost level of bubbles is known (and is as in Fig. 2), might conjecture that the quantum state corresponding
because we have specified it using explicit cosmological to null boundary conditions is simpler. In general, how-
boundary conditions. One might argue that this scenario ever, there seems to be no strong argument against such
would have yet simpler boundary conditions than the sce- a boundary condition surface as compared to a null sur-
nario in which the inflaton is placed in the false vacuum face.

VII. SUMMARY AND CONCLUSIONS on the manifold. We, and others, have studied this iden-
tification on de Sitter space classically and quantum-
We have investigated the possibility of making “future- mechanically. Antipodally identified (or “elliptic”) dS
eternal” inflation eternal also to the past. Starting with has the virtues that it is causally stable and observers
a de Sitter spacetime background dominated by the false- have no event horizons. It has the disadvantage that its
vacuum energy in which true-vacuum bubbles can form, non-time-orientability makes defining a reasonable quan-
we specify that the bubble distribution at any time is tum field theory difficult. With antipodal identification
in exactly the steady-state configuration asymptotically our model is more economical as the two duplicate uni-
approached by semi-eternal inflation. All bubbles are verses are identified; however not all of the attractive
then equivalent, and the statistical distribution of bub- features of “pure” edS remain.
bles admits both a time-translation and space-translation Our model can be generalized to other inflaton po-
invariance in the background inflating space. tentials (such as for chaotic inflation), and therefore al-
The described region (“region I”) has no initial time, lows one to partially understand the global structure of
but has a null boundary J − that is the limiting surface the eternally inflating spaces that result. One may also
as t → −∞ of the flat spatial sections. The steady- use an analogous construction to specify a semi-eternally
state configuration at t > −∞ corresponds to the infla- inflating but non-singular universe by placing boundary
ton field being in the false vacuum state everywhere on conditions on a spacelike section.
J − , with no incoming radiation from J − . This sur- Our primary conclusion is that it is possible, using only
face can be reached by a past-directed geodesic of finite “standard” ingredients underlying popular models for in-
proper length from any point in the space, so region I is flation, to specify simple cosmological boundary condi-
geodesically incomplete; this is the “singularity” pointed tions on an infinite null surface that lead to past- and
to by theorems purporting that past-eternal inflation is future-eternal inflation. Such a universe would obey the
impossible [3, 4]. But the space can (and should) be ex- same Perfect Cosmological Principle that governs a semi-
tended, as the state on J − also constitutes boundary con- eternally inflating universe long after its beginning. If our
ditions for the region past J − if the manifold is extended. construction survives scrutiny, and can be specified at the
These boundary conditions imply that a duplicate copy fully quantum (or quantum-gravitational) level within a
of region I exists on the other side of J − . Together re- theory of fundamental physics, it could serve as the basis
gion I and the new “region II” constitute a geodesically for a realistic cosmology that avoids a cosmological sin-
complete (i.e. non-singular) inflating cosmology that ad- gularity, a beginning of time, or a creation of the universe
mits a coordinatization in which the background space “from nothing”.
and bubble distribution are time-translation and space- Acknowledgments
translation invariant.
The null surface J − constitutes a cosmological bound- We thank Alex Vilenkin for helpful comments on a
ary condition surface on which the universe is classically draft of this paper. AA is supported by a grant from the
in a particularly simple and symmetric state. Although W.M. Keck foundation. SG is supported in part by US
J − is an infinite null surface, some of its points are in Department of Energy grant DE-FG02-91ER40671.
causal contact, and one might attribute a finite invariant
volume of O(H −3 ) to it. Therefore specifying boundary
conditions on it is not like specifying them on an infi-
Here we review Green functions and the initial value
nite spacelike hypersurface (which would lead to a severe
problem in some detail, focusing for simplicity on the
“horizon problem”).
inhomogeneous scalar wave equation in fixed background
It might be possible to construct a quantum state cor- spacetimes. We are interested in how much information
responding to our classical boundary conditions on J − by must be specified, and where, in order to fix the field
taking a null-limit of spacelike sections on which the wave throughout spacetime.
functional describing the fields is centered on the desired
classical state. But an explicitly null quantum formula-
tion of our null boundary proposal has not been provided 1. Defining Green functions
and would constitute an interesting future study.
It is widely thought that the “arrow of time” is con-
Imagine that our scalar field φ satisfies
nected with cosmological boundary conditions. In our
model, time flows away from J − , and the AOT is con-  − m2 φ = q (xµ ) ,

sistent among all physical observers that can compare it.
The AOT is not, however, defined globally, and in our where q is an arbitrary source term, and we wish to de-
model the statistical description of the universe admits a termine φ at some point P with coordinates xP . This is
global symmetry that includes time-reversal. possible given φ and its time derivative on some complete
This symmetry, along with the presence of two dupli- spacelike slice through the past light-cone (PLC) of P .
cate, non-communicating universes, motivates—though Let us see why this is so, and whether this is the only
does not require—formally identifying antipodal points suitable set of initial data for the problem.

FLC or the PLC of P . The usual choice, consistent with

our standard ideas about causality, is to pick a spacelike
surface of constant time (see left side of Fig. 5). Then
evaluation of Eq. (13) requires φ and φ̇ on S. For a more
general ∂V , we require φ and its normal derivative on S.
V (Note that this surface need not be everywhere spacelike,
but if not then the data must be self-consistent.) The
case of present interest is that of a null surface, for which
the normal lies everywhere within the surface itself (this
S L is made possible by the Lorentzian signature of the space-
time.) Then specifying φ on the surface also specifies the
normal derivative of φ on it. A suitable null surface is,
for example, the forward light-cone L of a point Q to the
past of P ; now we need only know φ on the surface S
where this FLC of Q is within the PLC of P (see right
FIG. 5: Possible choices of V for different Green functions. side of Fig. 5). (If the data near Q is suitably regular, the
The solid lines indicate the regions S of ∂V which contribute non-smooth nature of the surface at Q is unimportant.)
to the integral. On the left, we have volumes suitable for In the same spirit, one might consider a “wedge” such as
the retarded and advanced Green functions GR and GA , and
the boundary of the future of the segment of a line in the
on the right we have volumes suitable for the commutator
function GC . past of P .
Let us now consider the second way of isolating φ(P ),
using the LHS of Eq. (13). Set s = 0, and choose some
First introduce another function G, which depends on piece of ∂V to be spacelike and run through P . Then
P and is assumed to satisfy choose G to vanish everywhere on this surface. Further,
choose it to have a δ-function at P in its normal deriva-
 − m2 G = s (xµ ) ,

(12) tive; the integral then picks up a contribution propor-
tional to φ(P ). This choice comprises boundary condi-
where s is to be chosen. Taking G times (11) minus φ tions for G, and the function they determine, which we
times (12) and integrating through some four dimensional denote GC , now vanishes outside the (full) light-cone of P
volume V yields and satisfies the homogeneous version of Eq. (12). Choos-
√ √ ing now the remainder of ∂V to close off to the future or
d4 x∂µ G −gg µν ∂ν φ − φ −gg µν ∂ν G

to the past of P , we again may determine φ(P ) once data
is given on a complete slice S through one of the light-

= −g (qG − sφ) d4 x. (13) cones. GC may be called the commutator function, be-
cause it turns out to be equal to −i times the commutator
of a free quantum field (at least in a globally hyperbolic
This allows determination of φ(P ) in terms of its values
spacetime). Note also that GC = GA − GR : subtract-
elsewhere by a suitable choice of s, G, and V.
ing the (inhomogeneous) equations governing GR and GA
To effect this, we must isolate φ(P ), either on the left
leaves a homogeneous equation for the difference. Then
or right hand side of (13). Let us first use the RHS. Take
the difference must vanish outside the light-cone of P ,
s to be a δ-function at P (or, more carefully, a function
just as for GC .
peaked near P that can be taken to a δ-function limit).
In all cases, G depends on the point P , and so may
Then choosing V to enclose P gives φ(P ) for the second
be considered as a function of two variables, x and xP .
term on the RHS of Eq. (13). Using Gauss’s law, this
Because φ satisfies (11), it turns out that the G also
equation can now be used to express φ(P ) as a surface
satisfies (11) with respect to xP , at least away from xP =
integral over of φ and its derivative over the boundary ∂V
of V, plus a volume integral over V of the source alone.
To fix G(x), we must choose boundary conditions spe-
cific to our choice of s. Two conventional choices are to 2. Green functions for Minkowski and dS
require G to vanish either to the future or to the past of spacetimes
P . In the first case G is known as the “retarded” Green
function and may be denoted GR ; in the second case it
Let us now outline a procedure to obtain some Green
is “advanced” (GA ).
functions for massive scalar fields in certain spacetimes,
We must now choose V (see Fig. 5). It turns out (see,
using a slightly unconventional but perhaps more intu-
e.g., [15]) that GA 6= 0 only on and within the FLC of
itive method not relying on the usual Fourier techniques
P , and that GR 6= 0 only on and within the PLC of P .
or iǫ-prescriptions. We start with the commutator func-
Hence only a segment S of ∂V contributes to Eq. (13),
tion for a massless field in 3+1 D Minkowski space,
enabling us to deduce φ(P ) using only data on some con-
GM 2 2

nected surface making a complete span through either the C = sgn (t) δ t − r
/2π, (14)

with coordinates such that P is at the origin. That coordinate patch of the open slicing of dS. The equation
this solves φ = 0 can be seen by expanding out the reads:
δ-function and comparing to a general superposition of
incoming and outgoing spherically symmetric waves. (A ∂t sinh3 Ht∂t φ + m2 φ = 0.

3 (17)
nice discussion of this function is found in [48].) Note sinh Ht
that this is only non-zero on the light-cone itself, vanish-
To solve this [50], write φ = Hχ/ sinh Ht and set z =
ing even inside the cone. This is a special property of
cosh Ht to obtain
massless fields in even-dimensional spacetimes (see [15]
for more details.) m2
To obtain the solution for the massive field in (1 − z 2 )χ,zz − 2zχ,z + 2 − 2 − χ = 0, (18)
H 1 − z2
Minkowski space, we start by considering
which is Legendre’s equation [49] with ν = −1/2 +
G1 ≡ f t2 − r2 sgn (t) Θ t2 − r2 .
9/4 − m2 /H 2 and µ = −1. The solution to this equa-
tion which is regular as z → 1 is Pνµ (z). In terms of t,
The sgn and Θ functions enforce the gross features of this tends to Ht/2 as t → 0 independent of m2 , hence φ
the behaviour that we require. Writing  using spherical tends to H/2.
polar coordinates, we find We can now use this result to deduce the form of the
commutator over all of dS space. Extend the meaning of
( − m2 )G1 = sgn (t) Θ t2 − r2  − m2 f
τ 2 to be the signed geodesic distance squared between P
− 4f (0) sgn(t)δ t2 − r2 .

(15) and the point in question. Note that we do not consider
the point to be in the light-cone of the antipode of P ,
Let us choose f (τ 2 ) (where here τ 2 = t2 − r2 ) such that since no geodesic exists which connects it to P . In this

 − m f = 0 inside the light-cones; a general solution case we rather define GC to be zero (but see our discus-
is f (τ 2 ) = AJ1 (mτ )/τ + BY1 (mτ )/τ where J1 and Y1 are sion of the antipodal quantum commutator above.) Near
Bessel functions as in Ref. [49]. The first term of Eq. (15) P the space is locally Minkowskian, so we can compare
is then zero. For the second term, we notice that it is to our Minkowksi result to obtain:
basically the solution to the massless problem that we’ve
already found. So let us simply add the massless solution
m2 Θ(H 2 τ 2 ) −1
GM M0 sgn(t)
C from Eqn. (14) to our ansatz G1 . Since GC = 0, δ(τ 2 ) −
C = Pν (cosh Hτ ) .
2 M0
we have left −m GC on the RHS. By choosing A = 2π 2 sinh Hτ
−m/4π and B = 0, we obtain a complete cancellation, (19)
leaving us with our desired solution. Thus the full Green Here by t is meant some suitable generalization of
function is Minkowski time, with sgn(t) thus serving to make GdS C
be of opposite
√ sign in the future and past light-cones, and
 mΘ(τ 2 )J1 (mτ ) τ = + τ 2 . The sgn(t)δ(τ 2 ) should be interpreted as the
sgn (t) 2
GMC = δ τ − . (16) generalization of the equivalent term in the Minkowski
2π 2τ
We may now consider generalizing Eq. (16) to other
spacetimes such as dS. First, let us note that we took
f above to be a function of the proper time τ from the 3. Domains of dependence
origin, within the light-cone. Thus the commutator func-
tion was invariant under Lorentz boosts. This suggests Given our Green functions, we would like to investigate
that in dS, for example, we make our commutator inside to what extent fixing the fields on a null-cone L of a
the forward light-cone invariant under isometries which point Q determines the field values within that cone. We
leave P fixed. This is the same as saying that it should be are particularly interested in the importance of the field
a function of the proper time from the origin alone. One near Q, as our cosmological boundary surface J − can
can define this quantity in terms of an “angle” in the 5D be considered the light-cone with Q “at” past-timelike
embedding space [11] (see footnote to Sec IV D 2), but infinity.
perhaps a more intuitive way to proceed is as follows. A Take, as a first example, the Green function GR for a
patch of dS can be covered by coordinates in which the massless field such as the scalar field in 4D Minkowski
metric is the same as that for an open expanding FRW space with GR from Eq. (14). Because GR has support
universe, with scale factor H −1 sinh Ht. The proper time only on the PLC of P , the field at P depends only on the
along the geodesics representing comoving observers is intersection of P ’s PLC and L. This indicates two things.
just given by the coordinate time t. These geodesics all First, specifying data on L explicitly determines the field
intersect at a point as t → 0, which we choose to be everywhere inside L (i.e. everywhere in the future of Q).
P , and they cover all of the interior of the FLC of P . Second, the field at any point P inside L does not de-
We therefore need only find an appropriate spatially ho- pend on φ(Q); we might thus consider Q to be irrelevant
mogeneous solution of the massive wave equation in the in terms of what occurs within L. This, however, hides

a subtlety: while Gr and φ on L always allow the con- the limit τ → ∞). More explicitly, we may consider a
struction of a valid solution of the field equations, there boundary surface comprised of J − at t̃ > t̃0 for some
is no guarantee that the field so obtained is continuous t̃0 , closed by the spacelike surface t̃ = t̃0 , where t̃ and
with the field specified on L unless some assumptions of t̃0 are in the global closed coordinates. Then the field
field regularity at the vertex (Q) are made: if we wish integral (13) consists of an integral along J − , and an
to evaluate φ on L to check that we have a genuine solu- integral over the t̃ = t̃0 hypersurface, which has a finite
tion to our boundary value problem, then we must either physical volume of order H −3 . But then as t̃0 → −∞,
know φ(Q), or assume that the field is regular as Q is τ → ∞, so this second contribution will vanish unless
approached along L, so that we can extend the field to either the average of the field or its t̃-derivative blow up
Q by continuity. faster than GR . Hence we expect that the field at any
For a massive field the strict “Huygen’s Principle” does point in the space will depend only on the values on J − ,
not hold; while much of the contribution to the integral and not on the field behavior at past infinity, as long as
in Eq.(13) comes from the PLC of P where GR is singular the fields are assumed to be suitably finite and regular.
(see [15] for some discussion of this “generalized Huygen’s This argument might also be applicable to gravity; the
principle”), there is also a contribution from inside P ’s equations governing the Weyl conformal tensor can be
PLC because GR is everywhere nonzero there. cast, at the linear level, as wave-equations for a spin-2
In the dS case, GR as given by Eq. (19) falls off expo- field, and boundary conditions can be specified on a null
nentially as τ → ∞; thus we may expect that when L is surface such as J − [24]. We therefore expect that our
taken to be J − , where the vertex lies “at” τ = ∞, the boundary conditions on J − determine the Weyl tensor
field values inside J − will not depend on the field at the uniquely, and therefore determine the spacetime to be
vertex (where by “at the vertex” we mean within J − in pure dS when bubble nucleations are neglected.

[1] A. Linde, D. Linde and A. Mezhlumian, Phys. Rev. D [18] See, e.g., H.D. Zeh, “The physical basis of the direction
54, 2504 (1996); A. Vilenkin, Phys. Rev. D 46, 2355 of time”, (Berlin: Springer-Verlag, 1999).
(1992). [19] R. Penrose, in “Quantum Gravity 2: A Second Oxford
[2] See, for example, A.H. Guth, Phys. Rept. 333, 555 (2000) Symposium”, ed. C.J. Isham, et al. (Oxford: Clarenden
and references therein; A.D. Linde, Phys. Lett. B175, Press), 1981.
395 (1986); A.D. Linde in The Very Early Universe, [20] G.F.R. Ellis and R. Maartens, qr-qc/0211082.
edited by G.W. Gibbons, S.W. Hawking and S.T.C. Sik- [21] R. Penrose, in “Fourteenth Texas Symposium on Rel-
los (Cambridge University Press, Cambridge, England, ativistic Astrophysics, Dallas, Tex., 1988” (New York:
1983); P.J. Steinhardt, same volume. Annals of the New York Academy of Sciences, 1989); S.
[3] For a review, see A. Borde and A. Vilenkin, Int. J. Mod. Hollands and R. Wald; gr-qc/0205058; N. Turok, Class.
Phys. D5 (1996) 813-824. Quantum Grav. 19, 3449 (2002).
[4] A. Borde, A.H. Guth and A. Vilenkin, gr-qc/0110012. [22] A. Linde, “Particle physics and inflationary cosmology”,
[5] A. Aguirre and S. Gratton, Phys. Rev. D 65 (2002) (Switzerland: Harwood Acedemic Publishers, 1990).
083507. [23] T. Vachaspati and M. Trodden, Phys. Rev. D 61, 023502
[6] A.H. Guth, Phys. Rev. D 23, 347 (1981). (2000).
[7] M. Bucher, A.S. Goldhaber and N. Turok, Phys. Rev. D [24] R. Penrose and W. Rindler, “Spinors and Space-Time,
52, 3314 (1995). Vol. I”, (Cambridge: Cambridge University Press, 1984).
[8] J.R. Gott, Nature (London) 295, 304 (1982); K. Ya- [25] T. Heinzl, Lect. Notes Phys. 572, 55 (2001),
mamoto, M. Sasaki, and T. Tanaka, Astrophys. J. 455, hep-th/0008096.
412 (1995); A. Linde and A. Mezhlumian, Phys. Rev. D [26] S. Gratton and N. Turok, Phys. Rev. D 63, 123514
52, 6789 (1995) (2001).
[9] A. Vilenkin, Phys. Rev. D 46, 2355 (1992). [27] S.W. Hawking and T. Hertog, Phys. Rev. D 66, 123509
[10] A.H. Guth and E.J. Weinberg, Nucl. Phys. B212, 321 (2002).
(1983). [28] S.W. Hawking and R. Penrose, The Nature of Space and
[11] For reviews of de Sitter space, see: Ref. [41]; M. Spradlin, Time (Princeton University Press, Princeton, New Jer-
A. Strominger, and A. Volovich, hep-th/0110007. sey, USA, 1996).
[12] S. Coleman and F. De Luccia, Phys. Rev. D 21, 3305 [29] A. Vilenkin, Phys. Rev. D 27, 2848 (1983).
(1980). [30] A. D. Sakharov, JETP Lett. 5, 24 (1967).
[13] B. Mandelbrot, The Fractal Geometry of Nature, (W.H. [31] A. D. Sakharov, Sov. Phys. JETP 52, 349 (1980).
Freeman and Co., New York, 1983). [32] A. D. Sakharov, Sov. Phys. JETP 56, 705 (1982).
[14] P.J. Steinhardt and N. Turok, hep-th/0111030. [33] A. D. Sakharov, Sov. Phys. JETP 60, 214 (1984).
[15] R. Courant, Methods of Mathematical Physics: Vol. II, [34] A. H. Guth, Phys. Rev. D 23, 347 (1981).
Partial Differential Equations (New York: Interscience [35] W. de Sitter, M.N.R.A.S 78,3 (1917).
Publishers, John Wiley & Sons, 1962). [36] E. Schrödinger, Expanding Universes (Cambridge Uni-
[16] R. Penrose, Gen. Rel. and Grav. 12, 225 (1980). versity Press, Cambridge, England, 1956).
[17] V.A.Rubakov and S.M.Sibiryakov Theor. Math. Phys. [37] H. Schmidt, Fortschr. Phys. 41, 179 (1993).
120, 1194 (1999). [38] G. Gibbons, Nuc. Phys. B271, 497 (1986).

[39] N. Sanchez, Nuc. Phys. B294, 1111 (1987); J.L. Fried- (1998).
man and A. Higuchi, Phys. Rev. D 52, 5687 (1995). [46] A. Linde, D. Linde and A. Mezhlumian, Phys. Rev. D
[40] M. Parikh, I. Savonije and E. Verlinde, hep-th/0209120. 49, 1783 (1994).
[41] S.W. Hawking and G.F.R. Ellis, The large scale structure [47] R. Bousso, Rev. Mod. Phys. 74, 825 (2002).
of space-time (Cambridge University Press, Cambridge, [48] P.A.M. Dirac, Lectures on Quantum Field Theory, (New
England, 1973). York: Belfer Graduate School of Science), 1996.
[42] F. Leblond, D. Marolf and R. C. Myers, JHEP 0206, 052 [49] M. Abramowitz and I. Stegun, “Handbook of Mathemat-
(2002). ical Functions” (New York: Dover Publications, 1972).
[43] E. Mottola, Phys. Rev. D 31, 754 (1985); B. Allen, Phys. [50] M. Sasaki, T. Tanaka and K. Yamamoto, Phys. Rev. D
Rev. D 32, 3136 (1985). 51, 2979 (1995).
[44] K. Lee and E.J. Weinberg, Phys. Rev. D 36, 1088 (1987).
[45] J. Garriga and A. Vilenkin 1998, Phys. Rev. D 57, 2230